Shafting vibration fault intelligent diagnosis method and system based on multi-source data fusion

Abstract

The invention belongs to the technical field of fault diagnosis, and discloses an intelligent shafting vibration fault diagnosis method and system based on multi-source data fusion. According to the invention, through multi-source distributed data fusion, part / parameter weight self-adaption, dynamic threshold setting and intelligent clustering fault analysis, the complex scene adaptability, diagnosis accuracy and system intelligence level of shafting vibration fault diagnosis are significantly improved; the defects of existing static clustering methods depending on FCM and the like in the aspects of complex fault recognition and multi-working-condition self-adaptability are effectively overcome, and the method has wide engineering application prospects.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to a method and system for intelligent diagnosis of shaft vibration faults based on multi-source data fusion. Background Art

[0002] Rotating machinery, as core power equipment in industries such as energy, electricity, and shipbuilding, faces a significant challenge. The stability of its shafting operation is directly linked to the overall safety and reliability of the equipment. According to statistics, approximately 63% of rotating machinery failures are caused by abnormal shafting vibration, including typical failure modes such as misalignment, rubbing, cracks, and looseness. Traditional shafting vibration monitoring relies primarily on time-frequency domain analysis using a single vibration sensor (such as an eddy current displacement sensor or accelerometer). This single vibration signal cannot fully reflect the dynamic characteristics of the shafting under complex operating conditions. This is especially true for early-stage, weak faults or coupled faults, where the signal-to-noise ratio is significantly reduced, making feature extraction difficult.

[0003] Chinese patent CN119779679A discloses a bearing fault diagnosis method and system. The method includes: collecting bearing vibration signals in real time during operation, dividing the vibration signals into multiple segments, and obtaining current, acoustic, and temperature data sequences corresponding to each vibration signal segment; calculating the noise level of each vibration signal segment; extracting the vibration characteristics of each vibration signal segment, as well as the corresponding current, acoustic, and temperature data sequence characteristics; obtaining a fused feature vector; clustering the fused feature vectors corresponding to each vibration signal segment using the FCM algorithm; and determining the bearing fault type based on the clustering results. However, existing methods rely on the FCM (Fuzzy C-Means) algorithm for cluster analysis to determine the fault type, which has limited ability to identify complex fault patterns and does not set abnormal reference thresholds based on dynamic factors such as component and parameter weights. This makes the fault diagnosis system lack sufficient adaptability to different operating conditions, component importance differences, and changes in monitoring data, and is prone to misjudgments or missed detections. To address these issues, we propose an intelligent shafting vibration fault diagnosis system and method based on multi-source data fusion. Summary of the Invention

[0004] The purpose of the present invention is to overcome the problems of existing methods that rely on the FCM (Fuzzy C-means) algorithm to perform cluster analysis to obtain fault types, have limited ability to recognize complex fault patterns, and do not set abnormal reference thresholds based on dynamic factors such as the weights of components and parameters. As a result, its fault diagnosis system lacks sufficient adaptability when facing different working conditions, differences in the importance of different components, and changes in monitoring data, and is prone to misjudgment or missed judgment. The present invention provides an intelligent diagnosis method and system for shaft vibration faults based on multi-source data fusion.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an intelligent diagnosis method for shaft vibration faults based on multi-source data fusion, comprising the following steps: A distributed monitoring topology model is established based on the shafting component topology. Multi-source component monitoring data is collected in real time through the distributed monitoring topology model and pre-processed. A reference database is constructed based on pre-processed multi-source component monitoring data, and weighted reference values ​​of components and corresponding parameters are set in the reference database. Abnormal reference thresholds are dynamically preset according to the weighted reference values ​​of components and corresponding parameters; Fuzzy clustering algorithm combined with abnormal reference threshold is used to analyze component monitoring data and build a fault analysis model; Perform weighted fusion on component monitoring data and output abnormal parameter sets; The abnormal parameter set is input into the fault analysis model, and the analysis model analyzes and judges the abnormal parameter set and outputs the fault diagnosis results including the fault device, fault type, and fault degree.

[0006] A further improvement of the present invention is that the specific method for real-time acquisition of multi-source component monitoring data through a distributed monitoring topology model is as follows: Obtain a device infrastructure diagram based on the distributed monitoring topology model, and construct a 3D device model based on the device infrastructure diagram; Set weight reference values ​​for equipment components in the equipment infrastructure diagram, and update the equipment 3D model according to the weight reference values ​​of the equipment components.

[0007] A further improvement of the present invention is that the specific method for preprocessing the multi-source component monitoring data is as follows: Acquire multi-source component monitoring data, preset the sampling frequency and upload frequency of the corresponding components in the multi-source component monitoring data, and set the sampling frequency and upload frequency to the device components in the three-dimensional model of the device; Identify the number of associated parameters and component weight reference values ​​of equipment components, determine the importance of equipment components based on the number of associated parameters and component weight reference values, and compare the importance of equipment components with a preset importance threshold; If the importance of the equipment component exceeds the preset importance threshold, the current equipment component is retained; If the importance of the equipment component does not exceed the preset importance threshold, the current equipment component in the equipment 3D model is deleted; Integrate the deleted 3D equipment model to complete the preprocessing of multi-source component monitoring data.

[0008] A further improvement of the present invention is that the specific method for dynamically presetting the abnormal reference threshold value according to the weight reference value of the component and the corresponding parameter of the component is as follows: Obtain reference values ​​of components and corresponding parameters of components and benchmark reference thresholds of multi-source component monitoring data; The analytic hierarchy process is used to determine the influence coefficients of the reference values ​​of components and their corresponding parameters on the benchmark reference thresholds of multi-source component monitoring data; The product value of the influence coefficient and the benchmark reference threshold is rounded up to obtain the threshold reference calibration value; The threshold reference coefficient corresponding to the threshold reference calibration amount is extracted, and the abnormal reference threshold is calculated using the threshold reference coefficient in combination with the benchmark reference threshold.

[0009] A further improvement of the present invention is that the technical method of influencing the coefficient is as follows:

[0010] in, represents the influence coefficient of the benchmark reference threshold, Indicates the current component The number of associated monitoring data types, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data.

[0011] A further improvement of the present invention is that the calculation method of the abnormal reference threshold is as follows:

[0012] in, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data, is the abnormal reference threshold, Indicates the threshold reference coefficient.

[0013] A further improvement of the present invention is that a fuzzy clustering algorithm is used in combination with an abnormal reference threshold to analyze component monitoring data. The specific method for constructing a fault analysis model is as follows: A search window is used to traverse the time-series multi-source component monitoring data, the search window is embedded based on an abnormal reference threshold, the search window is shifted, and at least one set of initial abnormal data is output to form an initial abnormal set; Obtaining an initial anomaly set, traversing at least one set of initial anomaly data in the initial anomaly set, clustering and analyzing the initial anomaly data based on a fuzzy clustering algorithm, obtaining anomaly clustering points of single-type anomaly data, and determining the Euclidean distance between the initial anomaly data and the anomaly clustering points; Perform negative correlation mapping on the Euclidean distance between the initial abnormal data and the abnormal cluster points to obtain the initial correlation between the initial abnormal data and the abnormal cluster points, and multiply the initial correlation with the influence coefficient to obtain the adjusted correlation; Calculate the mean of the initial correlation between the initial abnormal data and the abnormal cluster points, and determine whether the adjusted correlation is less than the mean of the initial correlation between the initial abnormal data and the abnormal cluster points. If it is less than the mean of the initial correlation between the initial abnormal data and the abnormal cluster points, determine that the corresponding initial abnormal data is an abnormal parameter.

[0014] A further improvement of the present invention is that the specific method of weighted fusion of component monitoring data and outputting an abnormal parameter set is as follows: Obtain at least one set of abnormal parameters, perform weighted fusion dimensionality reduction processing on the abnormal parameters based on the wavelet transform algorithm, and output the abnormal parameter set.

[0015] A further improvement of the present invention is that the specific method of analyzing and judging the abnormal parameter set by the analysis model is as follows: Obtain an abnormal parameter set, perform feature extraction on the abnormal parameters, and obtain a feature extraction result; Perform convolution fusion processing on the feature extraction results, combine the cuckoo algorithm to iteratively search the feature factor weights, aggregate local abnormal feature factors and global abnormal feature factors, and add faulty devices associated with the local abnormal feature factors and global abnormal feature factors; The local abnormal characteristic factors and the global abnormal characteristic factors are fused and processed to calculate the fault degree corresponding to the faulty equipment. Based on the corresponding fault type in the reference database based on the fusion processing of the local abnormal characteristic factors and the global abnormal characteristic factors, the faulty equipment, fault type and fault degree are integrated into the fault diagnosis result, and the fault diagnosis result including the faulty equipment, fault type and fault degree is output.

[0016] In a second aspect, the present invention provides an intelligent diagnosis system for shaft vibration faults based on multi-source data fusion, comprising: A monitoring data acquisition module is used to establish a distributed monitoring topology model based on the shafting component topology, collect multi-source component monitoring data in real time through the distributed monitoring topology model, and pre-process the multi-source component monitoring data; A dynamic threshold module is used to build a reference database based on pre-processed multi-source component monitoring data, set weight reference values ​​of components and their corresponding parameters with reference to the database, and dynamically preset abnormal reference thresholds based on the weight reference values ​​of components and their corresponding parameters; Model building module, used to analyze component monitoring data using fuzzy clustering algorithm combined with abnormal reference thresholds to build a fault analysis model; The weighted fusion module is used to perform weighted fusion on component monitoring data and output an abnormal parameter set; The fault diagnosis module is used to input an abnormal parameter set into the fault analysis model. The analysis model analyzes and judges the abnormal parameter set and outputs a fault diagnosis result including the fault device, fault type, and fault degree.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention fully utilizes dynamic monitoring data from different components and multiple channels by introducing multi-source data fusion and distributed monitoring topology, and weightedly fuses multivariate data to achieve a more accurate and comprehensive characterization of the operating status of each component of the shaft system, significantly improving the accuracy of identifying complex and atypical fault modes. Based on a reference database, the present invention assigns weights to each component and its corresponding parameters, and dynamically adjusts the abnormal reference threshold, so that the fault judgment threshold adapts in real time to the equipment operating status, working conditions and component importance, significantly enhancing the system's adaptability to variable working conditions and the robustness of diagnosis. The present invention utilizes weighted fusion and dynamic threshold setting, combined with a fuzzy clustering algorithm to analyze abnormal parameter sets, effectively suppressing the misleading effect of single or occasional abnormalities on the overall diagnostic results. The weight threshold is automatically increased for key components and sensitive parameters, and the diagnostic sensitivity is reduced for non-key components, thereby achieving targeted and hierarchical fault judgment, significantly reducing misjudgments and missed judgments caused by component characteristics, sampling errors or data anomalies, and improving the diagnostic reliability of the overall system. The present invention constructs a distributed monitoring topology model based on the topology of the shaft system components, which can adapt to monitoring nodes of different structures, different numbers and distribution forms, has good scalability and compatibility, facilitates subsequent functional expansion and cross-platform deployment, and provides a solid foundation for subsequent large-scale equipment monitoring and intelligent operation and maintenance. By inputting the abnormal parameter set after weighted fusion into the intelligent fault analysis model, the present invention can not only accurately locate the faulty component and fault type, but also quantify and output the degree of fault, meeting the requirements of intelligent, quantitative and refined fault diagnosis in actual engineering, and improving maintenance efficiency and equipment safety. In summary, the present invention significantly improves the adaptability to complex scenarios, diagnostic accuracy and system intelligence level of shaft system vibration fault diagnosis through multi-source distributed data fusion, component / parameter weight adaptation, dynamic threshold setting and intelligent clustering fault analysis, effectively overcomes the shortcomings of existing static clustering methods such as FCM in complex fault identification and multi-working condition adaptability, and has broad engineering application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the present invention; Figure 2 is a system diagram of the present invention; Figure 3 A schematic diagram of a process for collecting component monitoring data from multiple sources in real time through a distributed monitoring topology model; Figure 4A flowchart illustrating a method for dynamically presetting abnormal reference thresholds based on weighted reference values ​​of components and parameters; Figure 5 A flowchart illustrating a method for analyzing component monitoring data based on a fuzzy clustering algorithm combined with an abnormal reference threshold. Figure 6 A schematic diagram of the process for implementing the fault analysis model to analyze and judge abnormal parameter sets; Figure 7 This is a structural diagram of the intelligent diagnosis system for shaft vibration faults based on multi-source data fusion. DETAILED DESCRIPTION

[0019] In order to further understand the content of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0020] See also Figure 1 , an intelligent diagnosis method for shaft vibration fault based on multi-source data fusion, including the following steps: S1, establish a distributed monitoring topology model based on the shaft system component topology, collect multi-source component monitoring data in real time through the distributed monitoring topology model, and pre-process the multi-source component monitoring data.

[0021] S2, building a reference database based on the preprocessed multi-source component monitoring data, setting weight reference values ​​of components and their corresponding parameters in the reference database, and dynamically presetting abnormal reference thresholds according to the weight reference values ​​of components and their corresponding parameters.

[0022] S3, uses fuzzy clustering algorithm combined with abnormal reference threshold to analyze component monitoring data and build a fault analysis model.

[0023] S4, performs weighted fusion on component monitoring data and outputs an abnormal parameter set.

[0024] S5 inputs the abnormal parameter set into the fault analysis model, and the analysis model analyzes and judges the abnormal parameter set and outputs the fault diagnosis results including the fault device, fault type, and fault degree.

[0025] See also Figure 2 , an intelligent diagnosis system for shaft vibration faults based on multi-source data fusion, including: The monitoring data acquisition module is used to establish a distributed monitoring topology model based on the shaft system component topology, collect multi-source component monitoring data in real time through the distributed monitoring topology model, and pre-process the multi-source component monitoring data.

[0026] The dynamic threshold module is used to build a reference database based on the preprocessed multi-source component monitoring data, set the weight reference values ​​of the components and their corresponding parameters with the reference database, and dynamically preset the abnormal reference threshold according to the weight reference values ​​of the components and their corresponding parameters.

[0027] The model building module is used to analyze component monitoring data using a fuzzy clustering algorithm combined with an abnormal reference threshold to build a fault analysis model.

[0028] The weighted fusion module is used to perform weighted fusion on component monitoring data and output an abnormal parameter set.

[0029] The fault diagnosis module is used to input an abnormal parameter set into the fault analysis model. The analysis model analyzes and judges the abnormal parameter set and outputs a fault diagnosis result including the fault device, fault type, and fault degree.

[0030] Example 1: This embodiment provides a method for collecting component monitoring data from multiple sources in real time through a distributed monitoring topology model. Figure 3 A schematic diagram of a method for collecting multi-source component monitoring data in real time through a distributed monitoring topology model is shown. The method for collecting multi-source component monitoring data in real time through a distributed monitoring topology model specifically includes: Step S101: Obtain a diagram of the equipment infrastructure, construct a 3D model of the equipment based on the diagram, assign weight reference values ​​to equipment components using principal component analysis, and update the 3D model based on these weight reference values. This process highlights the importance of key components within the overall equipment, allowing the monitoring system to focus more specifically on these components. It also optimizes the 3D model to better reflect the key characteristics and operational priorities of the equipment. Step S102: Preset the sampling frequency and upload frequency of the corresponding component monitoring data based on the reference value of the equipment component weight. The sampling frequency and upload frequency of the component monitoring data are assigned to the equipment components in the equipment 3D model. By reasonably setting the sampling and upload frequencies, the timeliness and integrity of the key component monitoring data are guaranteed, so that fault diagnosis can be analyzed and judged based on the latest and most sufficient data, thereby improving the timeliness and accuracy of fault diagnosis. Step S103: Identify the number of associated parameters and component weight reference values ​​for the equipment components, and use the entropy weight method to determine the importance of the equipment components. This comprehensive consideration of multiple factors allows for a more comprehensive and objective assessment of the importance of equipment components within the entire system, avoiding the bias that can result from single-factor evaluations. Step S104: determining whether the importance of the equipment component exceeds a preset importance threshold. In this embodiment, the importance threshold may be 0.1-0.15. Step S105: If the importance exceeds the preset threshold, the current device component is retained; Step S106: if the importance of the equipment component does not exceed the preset importance threshold, delete the equipment component from the equipment 3D model; Step S107 , integrating the deleted three-dimensional equipment models to obtain a distributed monitoring topology model. The distributed monitoring topology model collects component monitoring data from multiple sources in real time based on the sampling frequency and upload frequency of the component monitoring data.

[0031] In this embodiment, the distributed monitoring topology model, derived from integrating and reducing the 3D equipment model, efficiently collects multi-source component monitoring data in real time based on the sampling and upload frequencies of component monitoring data. This model construction approach makes the monitoring network more efficient and rational, enabling rapid and accurate collection of key component operating data, providing timely and reliable data support for subsequent fault diagnosis.

[0032] In an embodiment of the present invention, when multi-source component monitoring data is collected in real time through a distributed monitoring topology model, key components are screened based on the weights and importance of the components in the three-dimensional model of the equipment to ensure that the collected data is closely related to fault diagnosis, thereby improving diagnostic accuracy. In addition, key components are accurately located through importance threshold judgment, so that they receive special attention in diagnosis, thereby improving the extraction effect of key component fault characteristics and enhancing diagnostic reliability.

[0033] Example 2: This embodiment provides a method for dynamically presetting abnormal reference thresholds based on weighted reference values ​​of components and parameters. Figure 4 A schematic flow chart of a method for dynamically presetting an abnormal reference threshold value based on weighted reference values ​​of components and parameters is shown. The method for dynamically presetting an abnormal reference threshold value based on weighted reference values ​​of components and parameters specifically includes: Step S201, obtain the weight reference values ​​of components and parameters and the benchmark reference threshold of monitoring data, and determine the influence coefficient of the weight reference values ​​of components and parameters on the benchmark reference threshold of monitoring data based on the hierarchical analysis method; the application of the hierarchical analysis method combines qualitative analysis with quantitative analysis, and can quantify the degree of influence of component and parameter weights on the benchmark threshold as a specific influence coefficient, providing key parameters for the subsequent accurate calculation of the abnormal reference threshold, thereby improving the scientificity and accuracy of fault diagnosis.

[0034] The influence coefficient is calculated by the following formula: (1) in, represents the influence coefficient of the benchmark reference threshold, Indicates the current component The number of associated monitoring data types, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data; Step S202: Load the influence coefficient of the monitoring data baseline reference threshold, and round up the product of the influence coefficient and the baseline reference threshold to obtain a threshold reference calibration amount. In an embodiment of the present invention, after loading the influence coefficient of the monitoring data baseline reference threshold, round up the product of the influence coefficient and the baseline reference threshold to obtain the threshold reference calibration amount. This method of calibrating the baseline threshold based on the influence coefficient fully considers the different effects of components and parameter weights, and can more reasonably adjust the baseline threshold. Step S203 : extracting a threshold reference coefficient corresponding to the threshold reference calibration amount based on the time weighted function, and calculating an abnormal reference threshold using the threshold reference coefficient in combination with the benchmark reference threshold.

[0035] In this embodiment, the abnormal reference threshold is expressed as: (2) in, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data, is the abnormal reference threshold, Indicates the threshold reference coefficient.

[0036] In an embodiment of the present invention, a threshold reference coefficient corresponding to the threshold reference calibration amount is extracted based on a time-weighted function, and the coefficient is used in combination with the benchmark reference threshold to calculate the abnormal reference threshold. The threshold can be dynamically adjusted according to the time factor, so that the abnormal reference threshold can reflect the changes in the operating status of the equipment in real time and better adapt to the dynamic operation process of the equipment. By introducing the time-weighted function, the importance differences of the monitoring data at different time points are fully considered, so that the abnormal reference threshold can more accurately characterize the fault characteristics of the equipment, thereby improving the accuracy and reliability of fault diagnosis, and facilitating timely and accurate detection and warning of equipment failures.

[0037] Example 3: This embodiment provides a method for analyzing component monitoring data based on a fuzzy clustering algorithm combined with an abnormal reference threshold. Figure 5The figure shows a flow chart of a method for analyzing component monitoring data based on a fuzzy clustering algorithm combined with an abnormal reference threshold. The method for analyzing component monitoring data based on a fuzzy clustering algorithm combined with an abnormal reference threshold specifically includes: Step S301: Load component monitoring data, use a search window to traverse the sequential component monitoring data, embed a search window based on an abnormal reference threshold, and shift the search window to perform a shift search of the component monitoring data. Output at least one set of initial abnormal data to form an initial abnormality set. By loading component monitoring data, using a search window to traverse the sequential component monitoring data, and embedding an abnormal reference threshold to perform a shift search, it is possible to quickly and efficiently filter out potential abnormal data from a large amount of monitoring data and output an initial abnormality set. This avoids analyzing all data one by one, improves data processing efficiency, reduces the amount of data for subsequent analysis, and enables the fault diagnosis process to focus more on potential abnormal situations. Step S302: obtaining an initial anomaly set, traversing at least one set of initial anomaly data in the initial anomaly set, clustering and analyzing the initial anomaly data based on a fuzzy clustering algorithm, obtaining anomaly cluster points of single-type anomaly data, and determining the Euclidean distance between the initial anomaly data and the anomaly cluster points; In step S303, a negative correlation mapping is performed on the Euclidean distance between the initial abnormal data and the abnormal cluster points to obtain the initial correlation between the initial abnormal data and the abnormal cluster points. The initial correlation is multiplied by the influence coefficient to obtain the adjusted correlation. By adjusting the correlation, those abnormal data that are critical to fault diagnosis can be highlighted, so that subsequent analysis can focus more on these important data, thereby improving diagnostic efficiency and accuracy and avoiding interference by a large amount of secondary abnormal data.

[0038] Step S304 calculates the mean of the initial correlations between the initial abnormal data and the abnormal cluster points, and determines whether the adjusted correlation is less than the mean of the initial correlations between the initial abnormal data and the abnormal cluster points. If so, the corresponding initial abnormal data is determined to be an abnormal parameter. The mean of the initial correlations between the initial abnormal data and the abnormal cluster points is calculated, and based on this, it is determined whether the adjusted correlation is less than the mean of the initial correlations, thereby determining whether the corresponding initial abnormal data is an abnormal parameter. This method can effectively screen out truly representative and diagnostically valuable abnormal parameters from the initial abnormal data, providing a more reliable basis for further fault analysis and reducing the possibility of misjudgment.

[0039] Step S305 , obtaining at least one set of abnormal parameters, performing weighted fusion dimensionality reduction processing on the abnormal parameters based on a wavelet transform algorithm, and outputting an abnormal parameter set.

[0040] In an embodiment of the present invention, screening based on abnormal reference thresholds can promptly detect fluctuations in component monitoring data that exceed the normal range, help to detect early signs of faults in advance, provide the possibility of taking timely measures, and reduce the risk of further deterioration of the fault. The use of fuzzy clustering algorithms to perform cluster analysis on the initial abnormal data can classify different types of abnormal data and obtain abnormal cluster points of single-type abnormal data. This helps to more accurately identify the type and pattern of abnormalities, provide more detailed and accurate information for subsequent fault diagnosis, and improve the pertinence and accuracy of diagnosis. Based on the wavelet transform algorithm, weighted fusion dimensionality reduction processing can be performed on it, and multiple abnormal parameters can be fused to extract key feature information and output an abnormal parameter set. This not only reduces the dimension and complexity of the data and improves the efficiency of data processing, but also can retain important information in the abnormal parameters through weighted fusion, making the abnormal parameter set more representative and general, which is conducive to subsequent fault analysis and diagnosis.

[0041] In an embodiment of the present invention, the fault analysis model is based on a convolutional neural network model. The convolutional neural network can automatically extract feature representations from low-level to high-level features from raw data. The combination of multi-layer convolution and pooling operations can capture complex patterns and structural information in the data, providing strong support for subsequent fault diagnosis. The model includes an input layer, a convolutional neural network architecture, and an output layer. The convolutional neural network architecture includes a feature extraction layer, a convolution layer, a hidden layer, and a global pooling layer. During pre-construction, the hidden layer is frozen and replaced by a feature fusion module. The feature fusion module can fuse features from different levels or sources, allowing the model to more comprehensively capture the feature information of abnormal parameters. By fusing local and global features, the model's ability to express fault features can be improved, thereby improving the accuracy of fault diagnosis. The cuckoo algorithm is introduced into the feature fusion module to iteratively search for feature factor weights. The cuckoo algorithm is an efficient global optimization algorithm that imitates the breeding behavior of cuckoos to find the optimal solution. It has the advantages of fast convergence speed and strong robustness. During the feature weight optimization process, the optimal weight combination can be quickly searched in the feature space to avoid falling into the local optimum. Among them, the feature extraction layer extracts the abnormal parameter features as follows: (3) in, represents the output representation of the feature extraction layer, Represent abnormal parameter input value and input mean respectively, Indicates the number of exception parameter types, is the weight of the feature extraction layer.

[0042] Example 4: The embodiment of the present invention provides a method for analyzing and judging abnormal parameter sets using a fault analysis model. Figure 6 The following is a schematic diagram of a method for analyzing and judging abnormal parameter sets using a fault analysis model. The method specifically includes: Step S401: Obtain an abnormal parameter set, encode the abnormal parameters in the abnormal parameter set at the input layer, and pass the encoded abnormal parameters to the feature extraction layer; Step S402: The feature extraction layer extracts features from abnormal parameters and maps the feature extraction results to the convolution layer. The convolution layer performs convolution fusion processing on the feature extraction results. The feature fusion module combines the cuckoo algorithm to iteratively search the feature factor weights, aggregate local abnormal feature factors and global abnormal feature factors, and add the faulty devices associated with the local abnormal feature factors and the global abnormal feature factors. Step S403: fuse the local abnormal characteristic factors and the global abnormal characteristic factors, calculate the fault degree corresponding to the faulty equipment, index the corresponding fault type in the reference database based on the fusion of the local abnormal characteristic factors and the global abnormal characteristic factors, integrate the faulty equipment, fault type, and fault degree into a fault diagnosis result, and output the fault diagnosis result including the faulty equipment, fault type, and fault degree.

[0043] In this embodiment, unlike traditional fully connected neural networks, convolutional neural networks greatly reduce the number of model parameters, reduce computational complexity, and improve model training speed through local receptive fields and weight sharing mechanisms. At the same time, the model's translation invariance to input data is enhanced. By optimizing feature weights through the cuckoo algorithm, the model's adaptability and generalization ability to different fault modes can be improved, enabling the model to have stronger diagnostic capabilities when faced with new fault data.

[0044] Example 5: The embodiment of the present invention also provides an intelligent diagnosis system for shaft vibration faults based on multi-source data fusion. Figure 7 The structure diagram of the intelligent diagnosis system for shaft vibration faults based on multi-source data fusion is shown. The intelligent diagnosis system for shaft vibration faults based on multi-source data fusion specifically includes: The monitoring data acquisition module 100 establishes a distributed monitoring topology model based on the shafting component topology, collects component monitoring data from multiple sources in real time through the distributed monitoring topology model, and pre-processes the component monitoring data; The dynamic threshold module 200 loads pre-processed component monitoring data, builds a reference database based on the component monitoring data, sets weight reference values ​​for components and parameters using principal component analysis, and dynamically presets abnormal reference thresholds based on the weight reference values ​​of components and parameters; The fault diagnosis module 300 analyzes the component monitoring data based on the fuzzy clustering algorithm combined with the abnormal reference threshold, performs weighted fusion on the component monitoring data, outputs an abnormal parameter set, takes the abnormal parameter set as input, executes a pre-built fault analysis model, and the fault analysis model analyzes and judges the abnormal parameter set, and outputs a fault diagnosis result including the faulty device, fault type, and fault degree.

[0045] It should be noted that the monitoring data acquisition module 100, the dynamic threshold module 200, and the fault diagnosis module 300 are connected through communication or DTU. The intelligent diagnosis system for shaft vibration faults based on multi-source data fusion provided in the present invention corresponds to the above-mentioned intelligent diagnosis method for shaft vibration faults based on multi-source data fusion, which will not be repeated here.

[0046] In this embodiment, the monitoring data acquisition module 100 includes: A topology model building unit 110 is configured to build a distributed monitoring topology model based on the topology of the shaft system components; A real-time acquisition unit 120 collects component monitoring data from multiple sources in real time through a distributed monitoring topology model; The preprocessing unit 130 is used to preprocess the component monitoring data collected in real time and upload the preprocessed component monitoring data to the dynamic threshold module.

[0047] It should be noted that the real-time acquisition unit 120 is embedded with a piezoelectric acceleration sensor, an eddy current displacement sensor, a thermocouple sensor, and a piezoresistive pressure sensor to acquire data in real time.

[0048] In the embodiment of the present invention, the fault diagnosis module 300 includes: The abnormal parameter extraction unit 310 analyzes the component monitoring data based on the fuzzy clustering algorithm combined with the abnormal reference threshold, performs weighted fusion on the component monitoring data, and outputs an abnormal parameter set; The fault analysis unit 320 takes the abnormal parameter set as input and executes a pre-built fault analysis model. The fault analysis model analyzes and judges the abnormal parameter set and outputs a fault diagnosis result including the fault device, fault type, and fault severity.

[0049] In summary, the present invention provides an intelligent diagnosis system and method for shaft vibration faults based on multi-source data fusion. In an embodiment of the present invention, a convolutional neural network model is used as the basic architecture of the fault analysis model, and the cuckoo algorithm is introduced in the model construction process, which enables the fault analysis model to more deeply explore the complex features in the data. The cuckoo algorithm iteratively searches the feature factor weights, which can better integrate local abnormal feature factors and global abnormal feature factors, thereby improving the accuracy of fault diagnosis and identifying more complex fault modes. The hierarchical analysis method is used to determine the influence coefficient of the weight reference value of the component and parameter on the monitoring data benchmark reference threshold, thereby realizing dynamic adjustment of the threshold. The dynamic threshold setting method can better adapt to changes in the operating status of the equipment and improve the accuracy and reliability of fault diagnosis.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. An intelligent diagnosis method for shaft vibration faults based on multi-source data fusion, characterized in that: The following steps are involved: A distributed monitoring topology model is established based on the shafting component topology. Multi-source component monitoring data is collected in real time through the distributed monitoring topology model and pre-processed. A reference database is constructed based on pre-processed multi-source component monitoring data, and weighted reference values ​​of components and corresponding parameters are set in the reference database. Abnormal reference thresholds are dynamically preset according to the weighted reference values ​​of components and corresponding parameters; Fuzzy clustering algorithm combined with abnormal reference threshold is used to analyze component monitoring data and build a fault analysis model; Perform weighted fusion on component monitoring data and output abnormal parameter sets; The abnormal parameter set is input into the fault analysis model, and the analysis model analyzes and judges the abnormal parameter set and outputs the fault diagnosis results including the fault device, fault type, and fault degree.

2. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 1 is characterized in that: The specific method for real-time collection of multi-source component monitoring data through a distributed monitoring topology model is as follows: Obtain a device infrastructure diagram based on the distributed monitoring topology model, and construct a 3D device model based on the device infrastructure diagram; Set weight reference values ​​for equipment components in the equipment infrastructure diagram, and update the equipment 3D model according to the weight reference values ​​of the equipment components.

3. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 2 is characterized in that: The specific method for preprocessing multi-source component monitoring data is as follows: Acquire multi-source component monitoring data, preset the sampling frequency and upload frequency of the corresponding components in the multi-source component monitoring data, and set the sampling frequency and upload frequency to the device components in the three-dimensional model of the device; Identify the number of associated parameters and component weight reference values ​​of equipment components, determine the importance of equipment components based on the number of associated parameters and component weight reference values, and compare the importance of equipment components with a preset importance threshold; If the importance of the equipment component exceeds the preset importance threshold, the current equipment component is retained; If the importance of the equipment component does not exceed the preset importance threshold, the current equipment component in the equipment 3D model is deleted; Integrate the deleted 3D equipment model to complete the preprocessing of multi-source component monitoring data.

4. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 1 is characterized in that: The specific method for dynamically presetting the abnormal reference threshold value based on the weight reference value of the component and its corresponding parameter is as follows: Obtain reference values ​​of components and corresponding parameters of components and benchmark reference thresholds of multi-source component monitoring data; The analytic hierarchy process is used to determine the influence coefficients of the reference values ​​of components and their corresponding parameters on the benchmark reference thresholds of multi-source component monitoring data; The product value of the influence coefficient and the benchmark reference threshold is rounded up to obtain the threshold reference calibration value; The threshold reference coefficient corresponding to the threshold reference calibration amount is extracted, and the abnormal reference threshold is calculated using the threshold reference coefficient in combination with the benchmark reference threshold.

5. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 4 is characterized in that: The technical method of the influence coefficient is as follows: in, represents the influence coefficient of the benchmark reference threshold, Indicates the current component The number of associated monitoring data types, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data.

6. The intelligent diagnosis method for shafting vibration faults based on multi-source data fusion according to claim 1 is characterized in that: The calculation method of abnormal reference threshold is as follows: in, The monitoring data types are Weight reference value, components The weight reference value, Indicates the benchmark reference threshold of monitoring data, is the abnormal reference threshold, Indicates the threshold reference coefficient.

7. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 1 is characterized in that: The specific method of using fuzzy clustering algorithm combined with abnormal reference threshold to analyze component monitoring data and construct a fault analysis model is as follows: A search window is used to traverse the time-series multi-source component monitoring data, the search window is embedded based on an abnormal reference threshold, the search window is shifted, and at least one set of initial abnormal data is output to form an initial abnormal set; Obtaining an initial anomaly set, traversing at least one set of initial anomaly data in the initial anomaly set, clustering and analyzing the initial anomaly data based on a fuzzy clustering algorithm, obtaining anomaly clustering points of single-type anomaly data, and determining the Euclidean distance between the initial anomaly data and the anomaly clustering points; Perform negative correlation mapping on the Euclidean distance between the initial abnormal data and the abnormal cluster points to obtain the initial correlation between the initial abnormal data and the abnormal cluster points, and multiply the initial correlation with the influence coefficient to obtain the adjusted correlation; Calculate the mean of the initial correlation between the initial abnormal data and the abnormal cluster points, and determine whether the adjusted correlation is less than the mean of the initial correlation between the initial abnormal data and the abnormal cluster points. If it is less than the mean of the initial correlation between the initial abnormal data and the abnormal cluster points, determine that the corresponding initial abnormal data is an abnormal parameter.

8. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 1 is characterized in that: The specific method for weighted fusion of component monitoring data and outputting abnormal parameter sets is as follows: Obtain at least one set of abnormal parameters, perform weighted fusion dimensionality reduction processing on the abnormal parameters based on the wavelet transform algorithm, and output the abnormal parameter set.

9. The intelligent diagnosis method for shaft vibration fault based on multi-source data fusion according to claim 1 is characterized in that: The specific method of analyzing and judging abnormal parameter sets by the analysis model is as follows: Obtain an abnormal parameter set, perform feature extraction on the abnormal parameters, and obtain a feature extraction result; Perform convolution fusion processing on the feature extraction results, combine the cuckoo algorithm to iteratively search the feature factor weights, aggregate local abnormal feature factors and global abnormal feature factors, and add faulty devices associated with the local abnormal feature factors and global abnormal feature factors; The local abnormal characteristic factors and the global abnormal characteristic factors are fused and processed to calculate the fault degree corresponding to the faulty equipment. Based on the corresponding fault type in the reference database based on the fusion processing of the local abnormal characteristic factors and the global abnormal characteristic factors, the faulty equipment, fault type and fault degree are integrated into the fault diagnosis result, and the fault diagnosis result including the faulty equipment, fault type and fault degree is output.

10. The intelligent diagnosis system for shaft vibration faults based on multi-source data fusion is characterized by: include: A monitoring data acquisition module is used to establish a distributed monitoring topology model based on the shafting component topology, collect multi-source component monitoring data in real time through the distributed monitoring topology model, and pre-process the multi-source component monitoring data; A dynamic threshold module is used to build a reference database based on pre-processed multi-source component monitoring data, set weight reference values ​​of components and their corresponding parameters with reference to the database, and dynamically preset abnormal reference thresholds based on the weight reference values ​​of components and their corresponding parameters; Model building module, used to analyze component monitoring data using fuzzy clustering algorithm combined with abnormal reference thresholds to build a fault analysis model; The weighted fusion module is used to perform weighted fusion on component monitoring data and output an abnormal parameter set; The fault diagnosis module is used to input an abnormal parameter set into the fault analysis model. The analysis model analyzes and judges the abnormal parameter set and outputs a fault diagnosis result including the fault device, fault type, and fault degree.

Images